fix moe params and run error

vllm/attention/ops/flashmla.py

@@ -27,7 +27,8 @@ if current_platform.is_cuda():
     except ImportError:
         _flashmla_extension_C_AVAILABLE = False
 else:
-    _flashmla_extension_C_AVAILABLE = False
+    _flashmla_extension_C_AVAILABLE = True
+    _flashmla_extension_C_AVAILABLE = True
     
 if current_platform.is_rocm():
     import flash_mla_cuda
@@ -42,7 +43,7 @@ def _is_flashmla_available() -> tuple[bool, str | None]:
             "compiled due to insufficient nvcc version or a supported arch "
             "was not in the list of target arches to compile for.",
         )
-    if not _flashmla_extension_C_AVAILABLE:
+    if not _flashmla_extension_C_AVAILABLE or not current_platform.is_rocm():
         return (
             False,
             "vllm._flashmla_extension_C is not available, likely "

vllm/envs.py

@@ -1747,7 +1747,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
     # vLLM will use optimized topk_softmax + renormalize
     "VLLM_USE_TOPK_RENORM":
         lambda:
-        (os.environ.get("VLLM_USE_TOPK_RENORM", "True").lower() in
+        (os.environ.get("VLLM_USE_TOPK_RENORM", "False").lower() in
                 ("true", "1")),
     # vLLM will use fused RMS + RoPE kernel
     "VLLM_USE_FUSED_RMS_ROPE":

vllm/model_executor/layers/fused_moe/fused_moe.py

@@ -670,7 +670,7 @@ def fused_moe_kernel(
     BLOCK_SIZE_N: tl.constexpr,
     BLOCK_SIZE_K: tl.constexpr,
     GROUP_SIZE_M: tl.constexpr,
-    SPLIT_K: tl.constexpr,
+    # SPLIT_K: tl.constexpr,
     MUL_ROUTED_WEIGHT: tl.constexpr,
     top_k: tl.constexpr,
     compute_type: tl.constexpr,
@@ -1185,7 +1185,7 @@ def zero_experts_compute_triton(
 
 # Adapted from: https://github.com/sgl-project/sglang/pull/2628
 def get_config_file_name(
-    E: int, N: int, dtype: str | None, block_shape: list[int] | None = None
+    E: int, N: int, dtype: str | None, block_shape: list[int] | None = None, use_nn_moe: bool | None = False,
 ) -> str:
     # device_name = current_platform.get_device_name().replace(" ", "_")
     # # Set device_name to H200 if a device from the H200 family is detected

vllm/model_executor/layers/fused_moe/unquantized_fused_moe_method.py

@@ -289,6 +289,8 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
         expert_load_view: torch.Tensor | None = None,
         logical_to_physical_map: torch.Tensor | None = None,
         logical_replica_count: torch.Tensor | None = None,
+        use_nn_moe: bool | None = False,
+        use_fused_gate: bool | None = False,
     ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
         if enable_eplb:
             assert expert_load_view is not None
@@ -316,6 +318,8 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
             expert_load_view=expert_load_view,
             logical_to_physical_map=logical_to_physical_map,
             logical_replica_count=logical_replica_count,
+            use_nn_moe=use_nn_moe,
+            use_fused_gate=use_fused_gate,
         )
 
     def get_fused_moe_quant_config(
@@ -351,10 +355,13 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
         expert_load_view: torch.Tensor | None = None,
         logical_to_physical_map: torch.Tensor | None = None,
         logical_replica_count: torch.Tensor | None = None,
+        use_nn_moe: bool | None = False,
+        use_fused_gate: bool | None = False,
     ) -> torch.Tensor | tuple[torch.Tensor, torch.Tensor]:
         topk_weights, topk_ids, zero_expert_result = layer.select_experts(
             hidden_states=x,
             router_logits=router_logits,
+            use_fused_gate=use_fused_gate,
         )
 
         if self.rocm_aiter_moe_enabled:
@@ -391,6 +398,7 @@ class UnquantizedFusedMoEMethod(FusedMoEMethodBase, CustomOp):
                 apply_router_weight_on_input=apply_router_weight_on_input,
                 global_num_experts=global_num_experts,
                 expert_map=expert_map,
+                use_nn_moe=use_nn_moe,
             )
 
         if layer.zero_expert_num != 0 and layer.zero_expert_type is not None:

vllm/model_executor/models/deepseek_v2.py

@@ -104,6 +104,7 @@ if current_platform.is_cuda_alike():
     from vllm import _custom_ops as ops
 elif current_platform.is_xpu():
     from vllm._ipex_ops import ipex_ops as ops
+from vllm.utils import W8a8GetCacheJSON
 
 
 logger = init_logger(__name__)
@@ -1540,7 +1541,7 @@ class DeepseekV2ForCausalLM(
         self.use_llama_nn = os.environ.get('LLAMA_NN') == '1'
         self.use_awq_pad = os.environ.get('AWQ_PAD') == '1'
         self.tritonsingleton= W8a8GetCacheJSON() 
-        self.tritonsingleton.topk = config.num_experts_per_tok
+        self.tritonsingleton.topk = self.config.num_experts_per_tok
         self.tritonsingleton.quant_method=self.quant_method 
 
     def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: